Perceived poor sleep quality in the absence of polysomnographic sleep disturbance in women with severe premenstrual syndrome

SUMMARY

Women with severe premenstrual syndrome (PMS) report sleep-related complaints in the late-luteal phase but few studies have characterized sleep disturbances prospectively. This study evaluated sleep quality subjectively and objectively using polysomnographic and quantitative electroencephalographic measures in women with severe PMS. 18 women with severe PMS (30.5 ± 7.6 y) and 18 women with minimal symptoms (controls, 29.2 ± 7.3 y) had polysomnographic recordings on one night in each of the Follicular and Late-luteal phases of the menstrual cycle. Women with PMS reported poorer subjective sleep quality when symptomatic in the Late-luteal phase compared with the Follicular phase (p < 0.05). However, there were no corresponding changes in objective sleep quality. Women with PMS had more slow wave sleep and slow wave activity than controls at both menstrual phases (p < 0.05). They also had higher trait-anxiety, depression, fatigue and perceived stress levels than controls at both phases (p < 0.05). Both groups showed similar menstrual phase effects on sleep, with increased spindle frequency activity and shorter rapid eye movement sleep episodes in the Late-luteal phase. In women with PMS, a poorer subjective sleep quality correlated with higher anxiety (r = −0.64, p = 0.005) and more perceived nighttime awakenings (r = −0.50, p = 0.03). Our findings show that women with PMS perceive their sleep quality to be poorer in the absence of polysomnographically-defined poor sleep. Anxiety has a strong impact on sleep quality ratings suggesting that better control of mood symptoms in women with severe PMS may lead to better subjective sleep quality.

INTRODUCTION

Premenstrual syndrome (PMS) is characterized by psychological and/or somatic symptoms that recur specifically during the luteal phase (~14 day period after ovulation) of the menstrual cycle and resolve by the end of menstruation, with continued absence during the follicular phase. Up to 18% of women have severe premenstrual symptoms that cause impairment and between 3 –10% of women meet diagnostic criteria for premenstrual dysphoric disorder (PMDD), a severe form of PMS, as defined in the DSM-IV (Halbreich, 2003). For a diagnosis of PMDD at least five of 11 severe symptoms must be present, of which one must be dysphoric-related. There is considerable variation in the duration of premenstrual symptoms, with some women experiencing symptoms for 2 weeks, and others for a few days, premenstrually.

One of the symptoms endorsable for a PMDD diagnosis is insomnia/hypersomnia and survey data indicate that women with PMS or PMDD report sleep-related complaints, such as frequent awakenings, difficulty initiating sleep after an arousal, fatigue, lethargy, and poor concentration (Mauri et al., 1988, Rivera-Tovar and Frank, 1990, Strine et al., 2005). Women with menstrual-related problems, including premenstrual syndrome, are at least two times more likely than other women to report insomnia and excessive daytime sleepiness, which interfere with their quality of life (Strine et al., 2005). However, few studies have characterized sleep-related disturbance prospectively in patients with PMS or PMDD using either subjective questionnaires or objective polysomnographic (PSG) recordings. Regarding subjective sleep assessments, some have found that women with severe PMS or PMDD report more awakenings and body movements (Mauri, 1990) or a poorer sleep quality (Baker et al., 2007) in the Late-luteal phase whereas others found no menstrual-phase differences in reported sleep time or in-bed wakefulness (Parry et al., 1995). It therefore remains unclear what aspects of the sleep experience are disturbed in PMS and how disturbed sleep is related to other premenstrual symptoms.

A limited number of studies have used PSG recordings to examine physiological sleep disturbances in women with PMS or PMDD (Baker et al., 2008). Sleep efficiency was found to be similar in women with PMS or PMDD in the asymptomatic Follicular phase and symptomatic Late-luteal phase (Baker, 2007 #350; Chuong, 1997 #17; Lamarche, 2007 #342; Lee, 1990 #16; Parry, 1989 #66; Parry, 1999 #18}. Interestingly, some trait-like differences in sleep architecture, evident during both follicular and luteal phases have been noted between women with PMS or PMDD and controls. However, specific findings are inconsistent across studies with reports of reduced slow wave sleep (SWS) and more Stage 2 sleep (Lee et al., 1990), less rapid eye movement (REM) sleep and more Stage 2 sleep (Parry et al., 1989), longer latency to REM sleep in women with PMS or PMDD (Baker et al., 2007), or no differences between groups (Parry et al., 1999). Altered sleep architecture, therefore, may characterize PMS/PMDD independent of symptom expression. As far as we are aware, a comprehensive comparison between subjective perceptions and objective measures of sleep has not been made in women with PMS at different menstrual phases.

A limitation of most previous studies in PMS is the lack of quantitative measures of sleep microstructure, such as EEG power spectral analysis, which provides a more sensitive indicator of underlying sleep disturbance than traditional sleep scoring. For example, patients with insomnia have higher beta EEG activity during sleep, reflecting greater cortical arousal, which may contribute to their sleep-state misperception (Krystal et al., 2002, Riemann et al., 2009).. In a previous study, we found some differences in the NREM sleep EEG, including a lower incidence of delta waves and a trend for more high-frequency spindle activity, in a group of women with PMS compared to controls, regardless of menstrual phase (Baker et al., 2007). To explore further the impact of PMS on sleep, the current study aimed to address the following questions: (1) do women with severe PMS perceive their sleep to be poorer in the symptomatic Late-luteal phase compared with the Follicular phase; (2) do subjective sleep perceptions correlate with objective sleep measures derived from the PSG and quantitative EEG analysis; (3) are subjective and/or objective sleep disturbances apparent in the Follicular and Late-luteal phases (i.e., are trait-like) or only in association with premenstrual symptoms (i.e., are state-like)?

METHODS

Subjects

The study was approved by the Institutional Review Boards of SRI International and Palo Alto Medical Foundation Research Institute. All participants provided written informed consent. Most participants were recruited from the community. About 20% of PMS participants were recruited from community outpatient clinics after referral from their physician. Screening and assessment procedures are fully described in (Sassoon et al., 2011). Inclusion criteria for both groups were regular menstrual cycles of between 24 – 35 days, regular sleep-wake schedules, no chronic medical conditions, or medication use over the previous 3 months (including hormonal contraceptives), and no current Axis I psychopathology determined from structured clinical interview. Only three subjects currently smoked (1 control and 2 PMS): two smoked occasionally and one smoked fewer than four cigarettes per day. The characteristics of the 18 women with PMS (n = 11 women with prospectively-confirmed PMDD and n = 7 women with prospectively-confirmed severe PMS) and 18 controls included in the final sample are shown in Table 1. From clinical interview, seven women in the PMS group were identified as having a history of at least one Axis I disorder. Five had a past history of a single episode of major depressive disorder, three of whom also had a past history of substance use disorder, and two others had a past history of an anxiety disorder. Twelve of the PMS women and none of the controls endorsed ‘severe’ symptoms of insomnia/hypersomnia premenstrually. Four other PMS women reported moderate insomnia/hypersomnia symptoms and the remaining two PMS women reported none of these symptoms.

Table 1.

Characteristics of participants

	Controls	Women with severe PMS
Sample size	18	18
Age (years)	29.2 (7.3)	30.5 (7.6)
Body mass index (kg.m−2)	22.5 (2.1)	23.7 (3.8)
Ethnicity (No. Caucasians)	9	13
Have children (No.)	5	7
Years of education	16 (2)	16 (2)
Menstrual cycle length (days)	28 (2)	29 (3)
Menarche (age)	12 (1)	12 (1)
Premenstrual dysphoric disorder (No.)1	0	11
Past history Axis I disorder (No.)	0	7c
Daily Symptom Ratings (Follicular phase)2	7.2 (8.2)	48.7 (21.9)a
Daily Symptom Ratings (Late-luteal phase)2	11.5 (10.4)	136.4 (38.8)a,b

Subject groups did not differ on any characteristic except past history of Axis I disorder and daily symptom scores. Values are means (SD) unless otherwise indicated.

¹According to DSM-IV diagnostic criteria;

²daily symptom scores are averaged for the follicular and late-luteal phases over two menstrual cycles; higher scores reflect more severe symptoms.

^adifferent from controls (ANOVA, group effect: F(1,32) = 167.7, p < 0.01);

^bdifferent from follicular phase (ANOVA, interaction effect: F(1,32) = 67.8, p < 0.01, n = 16 controls);

^cFisher’s exact test, p = 0.008.

Screening Procedures

Participants completed the Penn Daily Symptom Rating Form (DSR), a validated measure of PMS (Freeman et al., 1996) over at least 2 menstrual cycles. Subjects rate 17 symptoms on a 5-point scale (0 = none to 4 = extreme). To qualify for severe PMS, women need to score 80 or greater on the DSR in their Late-luteal phase and show an increase of at least 50% from the Follicular phase score. To meet DSM-IV criteria for PMDD, women have to rate at least 5 PMDD symptoms as severe (3 or 4) on at least 2 premenstrual days, with the same symptoms being rated as absent or minimal (0 or 1) postmenstrually. Women with PMS had higher daily symptom ratings in the Late-luteal phase compared with the Follicular phase and compared to controls (Table 1). During both the screening and recording phases, women used a commercially available self-testing kit that detects the presence of luteinizing hormone in urine (One Step Ovulation Test, IND Diagnostic, Inc, CA) to confirm ovulation. Finally, subjects had an adaptation/screening night with clinical PSG to confirm the absence of a clinically-significant sleep disorder.

Study Procedures

Following their adaptation/screening night, participants had laboratory recordings on 2 occasions during their menstrual cycle: once during the mid-follicular phase (6–11 days after the onset of menstrual flow) and once during the Late-luteal phase (9–13 days after the luteinizing hormone surge). Women entered the study at different phases of the menstrual cycle; 5 women with PMS and 7 controls had their Late-luteal phase recording first. Women were requested to maintain their customary bedtime and wake-time schedules for at least 5 days before a study night. On study days, the women were requested to refrain from drinking caffeinated beverages after 15:00, not to drink any alcoholic beverages, and not to take naps. All participants registered 0 on the breathalyzer and tested negative on a urine drug test at each recording. Participants slept in sound-attenuated bedrooms where the ambient temperature was maintained between 20 °C and 22 °C. Lights-out and lights-on times were self-selected by participants.

Subjective Evaluations

Before going to bed, women completed the following questionnaires: Quality of life enjoyment and satisfaction questionnaire – short form, a validated measure of quality of life over the past week (Endicott et al., 1993); Perceived Stress Scale, a 10-item measure of the degree to which situations over the past week are appraised as stressful (Cohen et al., 1983); and the Profile of Mood States, a validated measure of current mood (how you are feeling right now). Within 1 hour of waking up in the morning, women completed a modified version of the Pittsburgh Sleep Diary; Spielberger State-Trait Anxiety Inventory, a validated measure of state and trait levels of anxiety; and the Beck Depression Inventory-II, a validated measure of depression over the past 2 weeks. As part of the morning sleep diary, participants rated their sleep quality on a 100 mm visual analogue scale with anchor points of “very bad” and “very good.” They also rated how restless their sleep was with anchor points of “very restless” and “not at all restless,” how refreshed they felt on awakening with anchor points of “not at all refreshed” and “very refreshed” and morning alertness, with anchor points of “not at all alert” and “extremely alert.”

Data acquisition and analysis

EEG, electrooculographic, and electromyographic recordings were made using E-series amplifiers and Profusion software (Compumedics, Abbotsford, Victoria, Australia) linked to appropriate transducers. Electrodes for EEG recordings were placed according to the international 10–20 system and cross-referenced to A1 or A2. EEG signals were digitized at a sampling rate of 256 Hz, high-pass filtered at 0.3 Hz, and low-pass filtered at 30 Hz. Thirty-second epochs were scored according to standard criteria by 2 scorers. Interrater reliability of sleep scoring was set at 0.90, and discrepancies were resolved by a third scorer. Total sleep time is the time spent asleep minus in-bed wakefulness during time in bed. Sleep efficiency is the percentage of total sleep time during time in bed. Sleep-onset latency is the time from lights-out to the first 3 epochs of any stage of sleep. The time between sleep onset and the first epoch of REM sleep is REM sleep-onset latency. Brief arousals (< 30 s) were scored using American Academy of Sleep Medicine criteria.

Quantitative EEG analysis of all-night NREM and REM sleep

Power spectral and period amplitude analyses were performed on central (C3-A2 or C4-A1) derivations using PASS PLUS EEG analysis software (Delta Software, St. Louis, Mo.). Spectral power reflects both the incidence and amplitude of waves in a particular EEG frequency band, whereas period amplitude analysis can determine the separate contributions of wave amplitude and incidence.

Data were refiltered with a bandpass of 0.15–60 Hz. A fast fourier transform routine was performed on 30 s epochs of 4 s Welch tapered windows with 2.0 s overlap, resulting in a frequency resolution of 0.25 Hz. Spectral power density (μV²/Hz) was calculated for the 0.3- to 1-Hz frequency bin and for subsequent 1-Hz frequency bins until 30 Hz. Spectral power density was also calculated for the main frequency bands: delta (or slow wave activity, 0.3 – <4 Hz); theta (4 – <8 Hz); alpha (8– <12 Hz); sigma (12 – <15 Hz); beta1 (15 – <23 Hz); beta2 (23 – <30 Hz). We also analyzed broad frequency bands with period-amplitude analysis (PAA) using zero-cross measures for delta or zero-derivative measures for all higher frequencies because these frequencies are often superimposed on slower waves and then do not cross zero voltage. The specific PAA measures investigated were “time in band” or “derivative time in band” (sum of all half-wave or derivative half-wave durations = wave incidence, measured in seconds) per 30-second epoch and “average sample amplitude” or “derivative estimate of sample amplitude” (measured in μV). Epochs containing low or high frequency artifact were excluded using a fully-automated process described in Feige et al., 1999. Power spectra and period amplitude values were averaged across NREM sleep (stages 2, 3, and 4) and REM sleep separately for the entire night. Power density values for each of the broad frequency bands were normalized with a logarithmic transformation. Quantitative EEG analysis could not be performed for one control subject due to excessive artifact.

Analysis of NREM-REM sleep cycles

NREM-REM sleep cycles were defined according to Feinberg and Floyd, 1979. NREM sleep episodes were initiated with Stage 2 sleep and contained at least 15 minutes of stage 2, 3, and 4 sleep. REM sleep episodes lasted at least 5 minutes except for the first episode which had no minimum duration requirement. Duration of the first three NREM-REM sleep cycles was assessed as all participants had at least three complete cycles. Delta and sigma power were calculated within the NREM sleep episodes of the first three cycles. Sleep cycle data from two control subjects could not be analyzed due to artifact.

Hormone analysis

Blood samples collected from subjects on each visit were analyzed for progesterone and estradiol using standard radioimmunoassay kits. The intraassay and interassay coefficients of variations were 8.8% and 9.7%, respectively, for the progesterone assay (Diagnostic Products Corporation), and 5.3% and 9.3%, respectively, for the estradiol assay (Diagnostic Systems Laboratories).

Statistical Analysis

Statistical procedures were performed using PASW Statistics 18 (SPSS Inc., Chicago, Illinois, USA). All results are reported as means (± SD). Visual analog scale measurements were normalized before statistical analysis through the arcsine square root transform. All subjective and objective PSG and all-night EEG measures were analyzed with repeated-measures 2-way ANOVAs at a 95% confidence interval, with “menstrual phase” as the within factor and “subject group” as the between factor. Differences in sleep cycles were investigated with repeated-measures 2-way ANOVAs with “menstrual phase” and “sleep cycle” as within factors. When Mauchly’s test of sphericity showed significance for cycle, degrees of freedom were adjusted using the Greenhouse-Geisser correction but original degrees of freedom are reported. Where significant interaction effects were found, t-tests were used to identify the origin of any differences. We examined correlations between subjective sleep quality and selected self-report and objective measures in each group of women in the Late-luteal phase using Pearson correlations. These factors were: perceived number of awakenings; perceived sleep onset latency; state anxiety; depression scores; objective sleep efficiency; NREM EEG delta power; NREM EEG beta power. We also examined correlations between subjective estimates and objective measures of sleep parameters in each group of women in the Late-luteal phase and used repeated-measures ANOVAs to determine whether any differences in subjective and objective measures differed between groups or according to menstrual phase. Finally, correlations between PSG measures delta and sigma power in NREM sleep and progesterone and estradiol levels were explored using Pearson correlations.

RESULTS

Hormone concentrations

Progesterone was higher [F (1,31) = 28.2, p < 0.001, n = 16 controls and 17 PMS] in the Late-luteal phase (Controls: 4.5±4.5 ng/ml; PMS: 6.0±5.2 ng/ml) than in the Follicular phase (Controls: 0.6±0.3 ng/ml; PMS: 0.8±0.4 ng/ml). There were no significant group or interaction effects. There were no significant effects for estradiol. Controls had estradiol levels of 75.4±32.4 pg/ml and 76.9±67.8 pg/ml in their Follicular and Late-luteal phases, respectively. Women with PMS had estradiol levels of 80.8±52.0 pg/ml and 75.6±46.4 pg/ml in their Follicular and Late-luteal phases, respectively.

Laboratory Subjective Measures

Subjective sleep ratings and ANOVA results are shown in Table 2. Women with PMS reported more awakenings, felt less refreshed, and tended to feel less alert in the morning than controls at both phases. Both groups perceived more awakenings in the Late-luteal phase than the follicular phase. There was a significant interaction effect for subjective sleep quality ratings: women with PMS had a poorer sleep quality in the Late-luteal phase than the Follicular phase (paired t-test, p = 0.03), whereas sleep quality did not differ between phases for Controls.

Table 2.

Subjective assessment of sleep following overnight laboratory recordings in the Follicular and Late-luteal phases of the menstrual cycle in women with severe PMS and controls.

	Controls (N = 18)		Severe PMS (N = 18)
Variable	Follicular phase	Late-luteal phase	Follicular phase	Late-luteal phase	ANOVA
Sleep onset latency (min)	17 (15)	16 (15)	26 (31)	27 (31)	Group: ns Phase: ns Interaction: ns
Number of awakenings#a	1.7 (1)	2.4 (2)*	2.5 (1)	3.2 (1)*	Group: F(1,33) = 5.1, p = 0.03 Phase: F(1,33) = 6.3, p = 0.02 Interaction: ns
Wakefulness after sleep onset (min)	19 (21)	18 (18)	21 (19)	24 (23)	Group: ns Phase: ns Interaction: ns
Sleep Quality (Very bad – Very good, mm)	63.0 (23.8)	69.2 (18.5)	66.2 (25.1)	54.3 (23.9)*	Group: ns Phase: ns Interaction: F (1,34) = 4.3, p = 0.04
Nighttime Restlessness (Not at all – Very, mm)	40.0 (26.5)	43.0 (23.4)	44.9 (30.0)	54.5 (24.2)	Group: ns Phase: ns Interaction: ns
Ease of waking (Not at all – Very, mm)	70.3 (15.3)	67.0 (20.5)	63.9 (32.1)	60.2 (27.8)	Group: ns Phase: ns Interaction: ns
Feeling Refreshed (Not at all – Very, mm)#	69.2 (17.1)	71.9 (17.4)	55.8 (26.8)	57.7 (21.1)	Group: F(1,34) = 4.9, p = 0.03 Phase: ns Interaction: ns
Alertness (Not at all – Very, mm)	73.5 (14.0)	72.4 (16.2)	60.1 (23.4)	58.4 (22.9)	Group: F(1,34) = 4.1, ns (p = 0.051) Phase: ns Interaction: ns

^*p < 0.05 compared with follicular phase;

^#p < 0.05 compared with controls;

^an = 17 controls

Mood-related measures are shown in Table 3. Women with PMS had a poorer quality of life, higher depression, trait-anxiety, and current mood scores, and greater levels of perceived stress and fatigue than Controls irrespective of menstrual cycle phase. There were also menstrual phase effects for quality of life, current mood state, perceived stress, and fatigue. Post-hoc t-tests revealed that women with PMS showed further deterioration in current mood state, quality of life, fatigue, and perceived stress in the Late-luteal phase compared to their own Follicular phase whereas these factors did not change across the menstrual phases in controls.

Table 3.

Variables (mean (SD)) derived from questionnaires completed by women with severe PMS and controls during laboratory visits in the Follicular and Late-luteal phases of the menstrual cycle.

	Controls (N = 18)		Severe PMS (N = 18)
Variable	Follicular phase	Late-luteal phase	Follicular phase	Late-luteal phase	ANOVA
Quality of Life	57.2 (6.3)	56.3 (6.0)	53.7 (5.7)	48.2 (7.2)*	Group: F (1,34) = 13.1, p = 0.001 Phase: F (1,34) = 5.3, p = 0.03 Interaction: F (1,34) = 3.0, ns (p = 0.09)
Beck Depression Inventorya	1.9 (2.8)	2.2 (3.4)	7.8 (3.1)	9.9 (5.6)	Group: F (1,33) = 35.0, p < 0.01 Phase: F (1,33) = 3.9, ns (p = 0.057) Interaction: ns
State Anxiety	31.6 (7.4)	34.3 (9.1)	37.9 (8.1)	37.7 (11.1)	Group: ns Phase: ns Interaction: ns
Trait Anxietya	32.7 (6.9)	32.4 (7.1)	39.2 (5.4)	39.2 (6.8)	Group: F (1,33) = 9.4, p = 0.004 Phase: ns Interaction: ns
Profile of Mood States	53.6 (11.6)	53.6 (9.0)	64.6 (17.8)	81.6 (24.0)*	Group: F (1,34) = 7.2, p = 0.01 Phase: F (1,34) = 18.4, p < 0.001 Interaction: F (1,34) = 7.3, p = 0.01
Perceived Stress	8.8 (5.4)	10.9 (7.5)	14.2 (5.3)	19.2 (5.3)*	Group: F (1,34) = 16.8, p < 0.01 Phase: F (1,34) = 10.8, p = 0.002 Interaction: ns
Fatigue	11.2 (2.1)	12.1 (2.6)	13.0 (3.5)	15.6 (3.7)*	Group: F (1,34) = 12.9, p < 0.01 Phase: F (1,34) = 6.9, p = 0.013 Interaction: ns

^*paired t-test, p < 0.05 compared with follicular phase;

^an = 17 controls; ns = not significant

Laboratory Polysomnography Measures

There were few differences in the PSG either according to group or menstrual phase (Table 4). Women with severe PMS had more SWS, less Stage 1 sleep, and fewer awakenings than controls at both menstrual phases. Controls had less Stage 1 sleep in the Late-luteal phase than the Follicular phase, an effect not seen in the PMS group. Women, particularly controls, tended to have more brief arousals in the Late-luteal phase than the Follicular phase (p = 0.051).

Table 4.

Variables (mean (SD)) derived from the polysomnograms of women with severe PMS and controls in the Follicular and Late-luteal phases of the menstrual cycle.

	Controls (N = 18)		Severe PMS (N = 18)
Variable	Follicular phase	Late-luteal phase	Follicular phase	Late-luteal phase	ANOVA
Total recording time (min)	459.1 (57.6)	466.2 (48.4)	453.8 (51.0)	471.7 (44.9)	Group: ns Phase: ns Interaction: ns
Total sleep time (min)	410.9 (58.3)	419.1 (55.8)	405.3 (53.0)	423.7 (51.1)	Group: ns Phase: ns Interaction: ns
Sleep efficiency (%)	89.5 (6.7)	89.9 (7.0)	89.2 (5.5)	89.9 (7.3)	Group: ns Phase: ns Interaction: ns
Sleep onset latency (min)	14.1 (15.4)	19.8 (21.4)	16.0 (12.2)	18.9 (19.0)	Group: ns Phase: ns Interaction: ns
Wakefulness after sleep onset (min)	34.0 (22.0)	30.4 (21.9)	32.5 (20.5)	28.8 (19.7)	Group: ns Phase: ns Interaction: ns
Latency to REM sleep (min)	78.7 (23.9)	77.3 (19.5)	96.7 (41.5)	77.4 (20.9)	Group: ns Phase: ns Interaction: ns
Stage 1 (%)#	6.5 (2.6)	4.9 (2.4)*	4.0 (1.5)	4.2 (1.8)	Group: F (1,34) = 7.1, p = 0.01 Phase: ns Interaction: F (1,34) = 4.9, p = 0.03
Stage 2 (%)	58.1 (6.8)	60.8 (4.4)	57.9 (6.6)	56.9 (5.0)	Group: ns Phase: ns Interaction: ns
Slow wave sleep (%)#	13.4 (6.4)	11.1 (4.7)	16.1 (6.3)	16.5 (5.2)	Group: F (1,34) = 6.2, p = 0.02 Phase: ns Interaction: ns
REM sleep (%)	22.3 (3.5)	23.2 (3.5)	22.0 (5.1)	22.4 (3.9)	Group: ns Phase: ns Interaction: ns
No. of Awakenings#	21 (10)	22 (12)	16 (7)	15 (5)	Group: F (1,34) = 4.6, p = 0.04 Phase: ns Interaction: ns
Arousal Index (No./h)	6.9 (2.8)	8.9 (3.9)	7.4 (3.4)	7.5 (2.4)	Group: ns Phase: F (1,34) = 4.1, ns (p = 0.051) Interaction: F (1,34) = 3.1, ns (p = 0.085)

^*paired t-test, p = 0.02 compared with follicular phase;

^#significant group effect; No. = number

Relationship between subjective and objective sleep measures

Subjective and objective measures of sleep onset latency (Controls: r = 0.58, p = 0.01; PMS: r = 0.94, p < 0.01) and wakefulness after sleep onset (Controls: r = 0.69, p = 0.002; PMS: r = 0.62, p = 0.006) were correlated in the Late-luteal phase. Both groups of women tended to overestimate sleep onset latency and underestimate wakefulness after sleep onset but these estimates did not differ between groups or according to menstrual phase (ANOVA P > 0.1 in all cases). Subjective and objective measures of number of awakenings were not correlated in either group.

In controls, subjective sleep quality in the Late-luteal phase was unrelated to objective measures of sleep efficiency, NREM EEG delta or beta power but was correlated with state anxiety (r = −0.50, p = 0.03) and depression ratings (r = −0.48, p = 0.049) and tended to correlate with perceived number of awakenings (r = −0.44, p = 0.067). In women with PMS, subjective sleep quality in the Late-luteal phase was also unrelated to any objective measures but correlated strongly with state anxiety (r = −0.64, p = 0.005) and perceived number of awakenings (r = −0.50, p = 0.034).

All-night quantitative sleep EEG measures

There were no significant group effects for any bands other than the delta band. Women with PMS had more delta power (Group: F(1,33) = 7.1, p = 0.012), associated with more time in the delta band (F(1,33) =4.1, p = 0.05) and waveforms of greater amplitude (F(1,33) =5.8, p = 0.02) during NREM sleep than controls in both menstrual phases.

Both groups of women showed a similar distribution of Late-luteal phase-Follicular phase spectral power differences, with a prominent peak in the range of sleep spindles (12 – 15 Hz) (Figure 1). Sigma power was greater (F(1,33) = 19.2, p < 0.001), associated with increased time in the sigma band (F(1,33) = 18.1, p <0.001) and waveforms of greater amplitude (F(1,33) =13.3, p = 0.001) in the Late-luteal phase than the Follicular phase in both groups of women. Alpha power was also higher (F(1, 33) = 4.5, p = 0.04) associated with a higher amplitude of alpha waveforms (F (1,33) = 8.9, p = 0.005) in the Late-luteal phase. Beta1 power was higher (F(1,33) = 14.1, p = 0.001), associated with increased time in the beta1 band (F (1,33) = 6.4, p = 0.016) and increased amplitude (F (1,33) = 12.0. p = 0.002) in the Late-luteal phase. More refined analysis of the 1-Hz frequency bands revealed that activity was higher in the 9–10 Hz band and in all 1Hz bands between 11 – 17 Hz, with the largest increase being in the 14 – 15 Hz bin in the Late-luteal phase compared with the Follicular phase (Figure 1). There were no significant group x phase interaction effects for any EEG frequency bands.

Figure 1.

Spectral EEG power in NREM sleep for women with severe PMS and controls during the Late-luteal phase of the menstrual cycle, expressed relative to values in the Follicular phase (horizontal line). Hz bins are identified by their lower boundary values. The horizontal bar at the bottom of the figure indicates significant menstrual phase effects (ANOVA, P<0.05).

There were no significant group, phase, or interaction effects for any frequency bands in REM sleep.

NREM-REM sleep cycles

Variables derived from analysis of the first three sleep cycles of the night are shown in Figure 2 and significant ANOVA results are shown in Table 5. There were no significant group effects for any variables apart from delta power, which was higher across NREM sleep episodes in women with PMS compared to controls. Both groups of women had shorter REM sleep episodes and higher sigma power in NREM sleep episodes in the Late-luteal phase than the Follicular phase. There were significant interaction effects for NREM sleep episode duration and NREM-REM sleep cycle duration, with women with PMS having shorter NREM sleep episodes, which contributed to their shorter sleep cycles overall in the Late-luteal phase compared with the Follicular phase. There was no difference in NREM sleep episode duration or sleep cycle duration between menstrual phases in controls. NREM sleep episodes became increasingly shorter and REM sleep episodes longer across sleep cycles in both groups. Delta power declined and sigma power increased across NREM sleep episodes in both groups. There were no significant cycle x phase interactions for any variables.

Figure 2.

Duration (mean ± SD) of NREM sleep and REM sleep episodes and delta and sigma power within NREM sleep episodes for the first three NREM-REM sleep cycles in 18 women with severe PMS and 16 controls in the Follicular and Late-luteal phases of the menstrual cycle. Absolute values of delta and sigma power are shown but statistical comparisons were made on log-transformed values. * p < 0.05, ANOVA phase or phase/group interaction effect. See text for details.

Table 5.

Significant ANOVA results for sleep cycle analysis in 18 women with severe PMS and 16 controls in the Follicular and Late-luteal phases of the menstrual cycle.

	Group	Phase	Cycle	Group x Phase
NREM-REM sleep cycle duration	ns	F(1,32) = 5.2 p = 0.03	ns	F(1,32) = 5.7 p = 0.02
NREM sleep episode duration	ns	Ns	F(2,64) = 4.1 p = 0.04	F(1,32) = 9.4 p < 0.01
REM sleep episode duration	ns	F(1,32) = 5.9 p = 0.02	F(2,64) = 38.3 p < 0.01	ns
Delta power in NREM episodes	F (1,32) = 4.4, p = 0.04	Ns	F(2,64) = 44.8 p < 0.001	ns
Sigma power in NREM episodes	ns	F(1,32) = 10.8 p < 0.01	F(2,64) = 6.1 p < 0.01	ns

ns = not significant

Relationship between sleep and progesterone and estradiol

In controls, higher levels of progesterone (r =−0.53, p = 0.03) and estradiol (r = −0.53, p = 0.03) were associated with a lower percentage of REM sleep (progesterone: r = −0.53, p = 0.03; estradiol: r = −0.53, p = 0.03) and a higher amount of wakefulness after sleep onset (progesterone: r = 0.48, p = 0.05; estradiol: r = 0.55, p = 0.02). Although tending in the same direction, neither progesterone (r = −0.37) nor estradiol (r = −0.32) levels correlated with %REM sleep in women with PMS. There was no relationship between wakefulness after sleep onset and hormone levels in women with PMS (p > 0.8). All-night NREM delta power was correlated with progesterone (r = 0.69, p = 0.003) and estradiol (r = 0.511, p = 0.043) in Controls but not in women with PMS (p > 0.1). All-night NREM sigma power did not correlate with progesterone and estradiol levels in either group (p > 0.1).

Discussion

The major finding of our study is that women with severe PMS experience poor subjective sleep quality in the absence of PSG or EEG indicators of sleep disturbance when they are symptomatic during the Late-luteal phase. These findings suggest that an aspect of the sleep process not measured with the PSG or quantitative EEG analysis is altered in PMS, impacting sleep quality assessments, and/or that non-sleep related factors determine subjective sleep quality assessments in PMS. Indeed, we have correlational evidence that anxiety plays a strong role in influencing sleep quality assessments in the Late-luteal phase.

We found evidence of trait-like differences in mood and sleep between women with PMS and controls. Women with PMS were more anxious, depressed, stressed, fatigued and were more likely to feel unrefreshed in the morning compared to controls even in the symptom-free Follicular phase. These symptoms worsened in the Late-luteal phase, which supports the hypothesis that women with PMS have an underlying vulnerability to the development of premenstrual symptoms that surface in response to a trigger in the luteal phase (Halbreich, 1997). Also, women with PMS had more SWS, more delta power and delta waveforms of higher amplitude in NREM sleep compared with controls in both phases. This finding is in contrast to our previous finding of less delta incidence in NREM sleep in women with PMS (Baker et al., 2007). The reason for discrepant results between our two studies is unclear. There may be subgroups of women with PMS who show different alterations in SWS compared with controls. We found that both groups of women showed the expected decline in delta power across sleep cycles but that delta power within the first three NREM sleep episodes was consistently higher in women with PMS than controls. These findings suggest that the decay in delta power across the night is normal but that some other aspect of sleep homeostasis is altered in women with PMS independent of symptom expression. Women with PMDD have a different recovery response to sleep deprivation, with an exaggerated enhancement of sleep efficiency and less Stage 1 sleep during recovery sleep compared to controls (Parry et al., 1999). These altered sleep responses may relate to the circadian rhythm disturbances that characterize this population (Shechter and Boivin, 2010).

Interestingly, patients with borderline personality disorder have higher delta power (Philipsen et al., 2005) or more Stage 4 sleep (Bastien et al., 2008) in addition to having poorer, nonrefreshing sleep and feeling fatigued upon awakening. Women with severe PMS or PMDD also have distinct personality traits, particularly higher neuroticism-related traits, such as trait-anxiety and stress susceptibility (Halbreich, 1997) and they are more likely to have traits of a personality disorder, particularly obsessive-compulsive disorder (Sassoon et al., 2011). Possibly, underlying personality factors may be associated with particular sleep EEG traits and may also influence how an individual perceives their sleep quality.

In addition to trait-like differences in sleep and mood between groups, we also found evidence of differences specific to the symptomatic Late-luteal phase (state-like). Women with PMS reported a poorer subjective sleep quality in association with premenstrual symptoms in the Late-luteal phase, which supports survey studies (Mauri et al., 1988, Rivera-Tovar and Frank, 1990, Strine et al., 2005) and our earlier laboratory experiment (Baker et al., 2007). Interestingly, there were no increases in perceived sleep onset latency or wakefulness after sleep onset in association with the perception of poorer sleep quality in women with PMS. Women perceived that they woke up more frequently in the Late-luteal phase than the Follicular phase but this finding was common to both PMS and control women and therefore not specific to PMS. This perception of increased number of awakenings was not matched by an increase in number of awakenings in the PSG in either group although there was a tendency for women, particularly controls, to have more frequent microarousals during sleep in the Late-luteal phase. Women with PMS and controls both underestimated time spent awake and overestimated sleep onset latency to a similar extent compared with objective measures, a common finding in healthy subjects (Baker et al., 1999). Women with PMS, therefore differ from insomniacs who are more inaccurate than good sleeper controls in their perceptions of sleep and wake times, underestimating their total sleep time relative to objective measures (Harvey and Tang, 2012). The misperception of sleep in women with PMS was specific to their overall sleep quality ratings. Women with PMS had high objective sleep efficiency (on average, 90 %), few awakenings, and a low arousal index during the night in the Late-luteal phase. They also had less Stage 1 sleep and more SWS than controls, which was apparent at both menstrual phases. A perception of poor sleep quality in the absence of objectively-measured poor sleep is not unique to severe PMS. Similarly, research has shown that patients with irritable bowel syndrome (Elsenbruch et al., 2002), borderline personality disorder (Bastien et al., 2008, Philipsen et al., 2005), and patients with insomnia (Harvey and Tang, 2012, Krystal and Edinger, 2008) may report poor sleep quality in the absence of objective sleep disturbances. In pregnant and postpartum women, subjective and actigraphy measures of sleep efficiency are not correlated and there is a strong association between poor subjective nighttime sleep, but not objective sleep, and greater psychological disturbances (Bei et al., 2010). Also, peri- and post-menopausal women report less satisfaction with their sleep yet have better polysomnographic sleep quality indices than premenopausal women (Young et al., 2003).

Quantitative analysis of the sleep EEG also did not reveal significant sleep disturbance in association with perceived poor sleep quality in women with PMS. In their Late-luteal phase, women with PMS showed increased high frequency activity, including in the beta1 range, in NREM sleep, which has been associated with a state of hyperarousal in insomniacs (Riemann et al., 2009). But, this effect was also apparent in controls and is a menstrual phase effect on sleep that has been shown previously (Baker et al., 2007, Driver et al., 1996). Peak activity was centered at 14–15 Hz in the upper sleep spindle frequency range and there was no increase in high frequency activity in REM sleep, a stage noted for the low incidence of sleep spindles, suggesting that the increased activity reflects increased spindles, which encroach into the alpha and beta1 bands. The only objective sleep measure that changed specifically in association with PMS symptom expression was NREM sleep episodes, which were shorter in the Late-luteal phase. Shorter NREM sleep episodes in addition to shorter REM sleep episodes that characterize the normal luteal phase may reflect less continuous sleep even though the composition of sleep and the sleep EEG are unaltered in women with PMS. Possibly, this change in sleep dynamics impacts subjective sleep quality. Further studies of the transitions between sleep stages and the ultradian sleep rhythm may be informative about sleep quality in PMS as they have in chronic fatigue syndrome (Kishi et al., 2008).

Objective sleep measures may not capture the complete sleep experience. For example, psychological state may influence sleep quality judgments by affecting the sleep appraisal process rather than sleep itself (Krystal and Edinger, 2008). Subjective feelings on the day following sleep are important factors for judging sleep quality (Harvey et al., 2008). Correspondingly, we found that sleep quality ratings in our subjects correlated with morning anxiety levels in the Late-luteal phase. A negative bias for sleep quality ratings has been indicated in patients with depression (Armitage et al., 1997), insomniacs (Edinger et al., 2000), and patients with irritable bowel syndrome (Elsenbruch et al., 2002). Our results suggest that, when experiencing symptoms, women with PMS may have a similar negative bias for sleep quality ratings. Our findings have implications for treatment of PMS. Cognitive-behavioral treatments that decrease premenstrual anxiety, for example, may help women control mood-related symptoms as well as help them to develop an ability not to generalize dysphoria to other aspects of life, including sleep. Cognitive-behavioral interventions have been shown to be helpful to women with PMS or PMDD (Rapkin, 2003).

As part of our study, we investigated menstrual phase effects on sleep architecture in relation to steroid hormone levels. REM sleep episodes were shorter in the Late-luteal phase, similar to previous findings (Driver et al., 2008). We also found that higher progesterone and estradiol levels in the Late-luteal phase correlated with less %REM sleep and more wakefulness in controls. Although not significant in women with PMS, the relationship between %REM sleep and hormone levels tended to follow the same direction as in controls. These findings support, in part, animal studies showing a reduction in REM sleep and NREM sleep after administration of estradiol and progesterone (Mong et al., 2011). Sex steroids may directly modulate sleep architecture through their effects on sleep regulatory systems (Hadjimarkou et al., 2008). Alternatively, REM sleep may be suppressed in the presence of a raised body temperature in the luteal phase.

It is hypothesized that progesterone metabolites acting on GABA_A receptors are responsible for the increased spindle frequency activity in the luteal phase (Driver et al., 1996). However, we found no correlation between progesterone and sigma activity in the Late-luteal phase in our sample. The pattern of NREM sleep sigma EEG activity in the luteal phase closely tracks that of body temperature (Driver et al., 1996) leading to the suggestion that raised body temperature contributes to the increased sigma activity (Deboer, 1998, Driver et al., 2008). The cause of the increased spindle frequency activity in the luteal phase remains to be determined.

Our findings should be considered in the context of the limitations of our study. Although the sample size is larger than for most previous studies about PMS and sleep, it is insufficient, given the diversity of symptoms in PMS, to investigate whether there are subgroups of women with particular premenstrual symptom profiles who may show substantial objective sleep disturbances in the Late-luteal phase. Another limitation is that we recorded sleep on only one night in the Late-luteal phase and we therefore cannot comment on whether there may be night-to-night variability in PSG sleep disturbances depending on the variability of premenstrual symptom severity. Finally, the sample studied here was not randomly selected and consisted mostly of non-treatment seeking women such that findings may not be generalized to all women with severe premenstrual syndrome.

In conclusion, we have shown that women with severe PMS experience a decline in subjective sleep quality but no associated change in objective measures of sleep in the symptomatic Late-luteal phase compared with the Follicular phase and controls. This perceived decline in sleep quality is related to anxiety emphasizing the strong impact that mood has on sleep quality ratings. Future studies could examine whether treatment of anxiety and other mood-related symptoms in severe PMS improves subjective sleep quality.